ka@hou3c.UUCP (Kenneth Almquist) (02/18/84)
This actually covers several different issues. First, relative naming vs. each machine having a unique name. Second, the specification of routes vs. the specification final destination only. Finally, if you have routes, the use of a host specific syntax vs. a single route syntax. I agree that machine names should be unique. The domain scheme pro- vides a way to do this. One problem with the scheme is that it was designed to handle naming on the Internet. Thus last I heard the plan was to have DARPA register register domains that met certain require- ments (including having two nameservers on the Internet). The result has been that people have gone ahead and announced the existance of domains like MAILNET and UUCP which are not registered anywhere, so name conflicts are possible. I also agree that it would be better if users did not have to specify routes, but it is difficult to avoid the necessity. What is worse is that the routes are currently buried into the domain specific string in many cases. For example, mail from CSNET has a return addresses like "spaf.gatech@CSNet-Relay.ARPA". My mailer will use an efficient route to reply to this, but only because of a hack I put in my mailer to rewrite this address as "spaf@gatech.csnet". Recently CSNet has been replacing the dot with a percent sign so that the code I wrote no longer works. I don't think that the necessity for providing explicit routes will disappear any time soon. There are lots of problems, depending on the domain (including, as I mentioned, the lack of even a registery of top level domains that don't talk to the Internet.) What I do hope is achievable in the relatively near term is the use of explicit routes as opposed to placing routing information in the domain specific string. With the former a mailer approach it is easy to determine the original address; with the latter it is necessary to know the peculiarities of particular hosts. Kenneth Almquist